Structured Data and Machine-Readable Signals for AI

A whole vocabulary of "machine-readable" tactics gets sold as the secret to AI visibility. Some help a little, one is unproven and widely misunderstood, and the honest summary is smaller than the hype. This chapter separates the low-cost real signals from the ones you should not pay for.

Does structured data help you get cited by AI?

Sometimes, and clearly so in one place: data that is hard to read from prose. Price, stock availability, and shipping details are awkward to state in flowing text but trivial to express in schema, and engines lean on that markup to extract them cleanly. For that kind of fact, structured data plainly helps. Structured data has a canonical technical home worth reading in full. Structured data →

Beyond that narrow case, the evidence is mixed. Studies conflict on whether structured data correlates with AI citations, and Google's own position is that schema is not required to appear in its AI features. The sensible read is to treat the benefit as plausible and low-cost rather than proven. Add schema because it wins rich results and helps engines understand your entities, not because someone promised it buys citations. If a vendor sells schema as a guaranteed AI lever, that is more confidence than the evidence supports.

Should you publish an llms.txt file?

You can, but do not expect it to move citations. llms.txt is a proposed markdown file at your site root that gives large language models a curated, link-based map of your most important content, meant to work around small context windows. Jeremy Howard of Answer.AI proposed it in September 2024. The spec is simple: an H1 with your site name as the only required element, an optional summary, and lists of markdown links. A companion llms-full.txt inlines fuller content.

The honest status is that it is proposed and not adopted. No major answer engine has confirmed reading llms.txt for ranking or citation. Google's John Mueller compared it to the keywords meta tag, a signal engines learned to ignore because anyone could write anything in it, and Google's own AI-features documentation says you do not need AI text files. It is low-cost to publish and some coding assistants do consume it, so it is fine to add for a developer-tool audience. Just do not sell it, or buy it, as a proven AEO lever. It is unadopted and unproven, and presenting it otherwise is exactly the overselling this Academy warns against. IndexNow and Bing →

The test for any machine-readable tactic is simple: has a major engine confirmed it reads the signal for citation? For schema on hard-to-read data, roughly yes. For llms.txt, not yet. Spend accordingly.

What actually makes your content machine-readable?

The unglamorous answer beats the exotic one: keep your meaningful content in server-rendered HTML text. Not inside images, not locked behind JavaScript that has to run before the words appear. AI crawlers render JavaScript far less reliably than Googlebot does, so client-only content is often simply invisible to them. There is no markup that rescues a page whose words never reach the crawler in the first place.

This is the single most important machine-readable decision you make, and it is a rendering choice, not a schema choice. Server-side rendering or static generation puts real text in the initial HTML where every crawler can read it, which is the same safe default that classic SEO recommends. Server-rendered text → Get the words in front of the machine first, then add markup to help it interpret them.

One quick test settles most arguments here. View the page source, not the rendered DOM in dev tools, and search it for the sentence you want an AI to quote. If the sentence is in that raw HTML, crawlers that do not run JavaScript can read it. If it only shows up after the page's scripts execute, you are betting your visibility on a rendering step that AI crawlers often skip. When the answer is in doubt, move the content into the server response and stop guessing.

So what is the honest summary?

Machine-readable AI visibility is not a separate science with its own secret files. It is standard good SEO plus quotability. Be crawlable and indexable. Keep substance in server-rendered text. Write self-contained, named, factual answers. Add structured data where it helps engines extract hard facts and where it earns rich results. Keep your entity facts consistent. That is the whole list.

Everything past that is polish, and some of it is noise dressed up as a lever. The cottage industry selling exotic files depends on you believing there is a hidden layer you are missing. There mostly is not. What there is, once your content is readable and quotable, is a real decision about which crawlers you let in and which you keep out. Controlling AI crawlers →